tmVar

tmVar extracts and normalizes sequence variant mentions from unstructured biomedical literature to map them to dbSNP RSIDs and integrate text-mined variants with curated genomic resources.

Key Features:

• CRF-based extraction: Uses a conditional random fields (CRF) text-mining approach to identify variant mentions from literature.
• Variant scope: Systematically identifies sequence variants at both protein and gene levels and adheres to Human Genome Variation Society (HGVS) nomenclature.
• Normalization to dbSNP: Maps extracted variant mentions to dbSNP Reference SNP cluster IDs (RSIDs) for standardized identifiers.
• Validation against curated databases: Cross-references variant-gene pairs with dbSNP annotations and ClinVar curations based on genomic positions.
• Performance: Achieves a high F-measure of approximately 90% for variant mention extraction.
• Large-scale application: Applied across the entirety of PubMed literature for broad coverage of reported variants.
• Novel variant identification: Identified 41,889 RS numbers associated with 9,151 genes not previously cataloged in ClinVar.
• Rare variant extraction: Extracted rare variants (MAF ≤ 0.01) across 3,849 genes for analysis of underrepresented, potentially deleterious variants.

Scientific Applications:

• Novel discovery identification: Enables detection of RS numbers and variant-gene pairs absent from ClinVar, supporting discovery of uncurated variants.
• Rare variant analysis: Provides datasets of rare (MAF ≤ 0.01) variants across thousands of genes for studies of deleterious and population-rare variation.
• Database enrichment and prioritization: Integrates text-mined variants with dbSNP and ClinVar to enrich curated resources and assist variant evaluation and prioritization.

Methodology:

Applies CRF-based natural language processing to extract variant mentions, normalizes mentions to dbSNP RSIDs following HGVS nomenclature, and validates mappings by cross-referencing variant-gene pairs with dbSNP and ClinVar using genomic positions across PubMed-derived literature.